ABSTRACT
This paper illustrates the advantages of using models with unobserved heterogeneity, with use of the heterogeneous Poisson process with covariates, in an application to wind turbines failure data. It is shown that inclusion of heterogeneity in the form of frailties (random effects) has more significant effect than introduction of available covariates, and individual frailties can substitute the effect of covariates, i.e. individual frailties can be used for indication of importance of a covariate. In presence of covariates, estimated individual frailties can behave as another independent covariate. It is also shown that individual frailties can provide additional important information, e.g. about spatial failure effects within one wind farm. Covariates inthis article are mainly used in the form of factor covariates which enables to analyze the effect of those covariates without any restrictive assumption about the functional form of such an effect.
